Pith. sign in

REVIEW 6 major objections 6 minor 26 references

Explaining Model Overfitting in CNNs via GMM Clustering

T0 review · 6 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that CNN overfitting can be detected by counting 'anomaly filters'—filters whose feature-map clusters contain outlier samples—and that pruning these filters improves validation accuracy.

desk verdict Plausible filter-level overfitting diagnostic, but the evidence is too inconsistent to support the claim; needs a fixed anomaly-filter definition and a re-run. read the letter →

arxiv 2412.10457 v1 pith:4T3Q6Q4I submitted 2024-12-12 cs.LG cs.AIcs.CV

Hui Dou , Xinyu Mu , Mengjun Yi , Feng Han , Jian Zhao , Furao Shen This is my paper · ORCID
classification cs.LGcs.AIcs.CV
keywords CNNinterpretabilityoverfittingGaussianMixtureModelanomalyfiltersfeaturemapclusteringfilterpruninggeneralizationCalinski-Harabaszindex
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that overfitting in convolutional neural networks has a measurable filter-level signature. By clustering each filter's feature maps with a Gaussian Mixture Model, the authors identify 'anomaly filters' that produce small clusters or outlier points. They argue that the number of such filters rises when a model overfits, that the outlier samples behind these filters exert unusually large gradients, and that masking the anomaly filters improves generalization on validation data. If correct, this gives a quantitative, architecture-agnostic indicator of overfitting that requires no labels or model modification.

What carries the argument

The central object is the anomaly filter, defined by three criteria applied per filter after projecting its feature maps to two dimensions via PCA and clustering with a Gaussian Mixture Model: (1) unbalanced class distribution, where small clusters or outliers form; (2) an abnormally high Calinski-Harabasz index relative to the layer average; and (3) sufficiently large activation values. The CH index supplies the quantitative filter-evaluation metric, and the GMM supplies the soft clustering that reveals rare outlier points. The number of clusters K is chosen dynamically for each filter by selecting the value that maximizes the CH index, and these machinery components together turn qualitative visual patterns into a computable filter-level score.

What would settle it

Train a CNN with strong regularization (e.g., dropout or weight decay) that prevents overfitting, and count anomaly filters across epochs: if the count stays high or rises without a corresponding drop in validation accuracy, the claimed link between anomaly filters and overfitting fails. Alternatively, induce overfitting by adding label noise to a dataset and verify whether anomaly-filter counts increase in exact step with validation loss; a mismatch would falsify the indicator's specificity.

Watch

Extended reading notes

Core claim

The paper's central claim is that anomaly filters—filters whose GMM-clustered feature maps contain small clusters or distant outlier points—are a reliable indicator of CNN overfitting. This is operationalized through three hypotheses: anomaly filters increase in number as training moves into the overfitting regime; the outlier samples defining these filters receive gradients several times larger than normal samples, driving the model to over-learn unusual details; and masking or pruning anomaly filters decreases training accuracy while increasing validation accuracy, thereby improving generalization. Evidence is presented across AlexNet, LeNet-5, and a simple three-layer CNN on CIFAR-10, CIFAR-100, and Fashion-MNIST, with tables showing higher anomaly-filter counts in overfitting versus well-trained models and mostly favorable validation-accuracy changes after single-filter masking.

Load-bearing premise

The load-bearing premise is that reducing each filter's feature maps to two dimensions with PCA and then clustering with a Gaussian Mixture Model produces small clusters whose member samples are genuinely 'outlier samples' that cause overfitting, rather than artifacts of the embedding or the clustering algorithm.

Editorial extensions

If this is right

  • If anomaly-filter counts track overfitting, then monitoring this count during training could serve as an early-stopping signal that does not require a separate validation set.
  • Pruning anomaly filters could become a post-hoc generalization-boosting step for trained CNNs, complementing or partially replacing regularization.
  • The method's architecture-agnostic nature suggests it could be applied to deeper or residual networks, where the paper notes residual modules add complexity but may reveal additional patterns.
  • The gradient disparity between outlier and normal samples implies that loss functions weighting or down-weighting outlier samples could reduce overfitting without pruning.
  • The filter-level evaluation could be combined with existing interpretation tools to flag specific channels responsible for memorizing atypical training examples.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The anomaly-filter count might generalize as a cheap proxy for the sharpness of the loss landscape, connecting the paper's findings to the sharp-minima literature the authors cite.
  • A testable extension would be to apply the same clustering pipeline to attention heads in vision transformers: if the phenomenon is general, heads with outlier-dominated clusters should appear in overfitting transformers as well.
  • The method's dependence on the two-dimensional PCA projection is a potential sensitivity point: a controlled experiment varying the projection dimension or using t-SNE/UMAP would reveal whether the anomaly-filter signal is an artifact of the embedding or a robust property of the filters.
  • The authors report that larger datasets were excluded due to too few anomaly filters; this suggests the phenomenon may be more pronounced in small-data regimes, implying a possible connection to dataset size that the paper leaves implicit.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

6 major / 6 minor

Summary. The paper proposes an unsupervised, filter-level method for interpreting CNNs: feature maps of each filter are projected to 2D via PCA and clustered with Gaussian Mixture Models (GMM). A filter is labeled an "anomaly filter" if its clusters are unbalanced, its Calinski-Harabasz index is abnormally high, and its activations are sufficiently large. The authors then propose three hypotheses: anomaly filters increase in overfitting models, outlier samples drive overfitting, and masking anomaly filters improves validation accuracy. Three experiments across AlexNet, LeNet-5, and a simple CNN on CIFAR-10, CIFAR-100, and Fashion-MNIST are presented as evidence. The paper claims the method is universally applicable to CNNs without modification.

Significance. If the central claims were supported, the paper would offer a label-free, filter-level overfitting indicator that could be useful for model interpretation and pruning. The approach is relatively simple and does not require modifying the network, and the use of quantitative clustering metrics is a strength. However, the evidence as presented is not convincing: the anomaly-filter definition changes across experiments, the third hypothesis is contradicted by the paper's own LeNet-5 results, and the experiments lack basic statistical safeguards such as multiple runs and error bars. The contributions are therefore not established at the level required for a journal publication.

major comments (6)
  1. [Section 4, Table 1, Section 4.3] The anomaly-filter definition is not stable across experiments: Table 1 sets λ=100, α=0.2, β=1, θ=0.2 for Experiment 1 but λ=50, α=0.2, β=1.2, θ=0.5 for Experiment 3, and Section 4.3 states that "stricter rules have been applied in this experiment" because masking too many filters reduces accuracy. Consequently, the filters counted in Table 2 are not the same population as those masked in Table 4, so the three experiments cannot jointly support the claimed overfitting indicator. A fixed, pre-specified threshold policy is required before the central claim can be evaluated.
  2. [Section 4.3, Tables 4 and 5] The third hypothesis, that discarding anomaly filters improves validation accuracy, is directly contradicted by the paper's own LeNet-5 results: Table 5 reports accuracy changes of -1.81% (CIFAR-10) and -1.70% (CIFAR-100) when anomaly filters are masked, and Table 4 shows validation accuracy increases in only 1/3 of LeNet-5 CIFAR-10 cases and 0/1 CIFAR-100 cases. The text softens the claim to "fewer accuracy drops compared to random maskings," which is a different, weaker statement than the one in the Introduction and Section 4.3. This inconsistency affects the core conclusion and must be resolved.
  3. [Section 4.1, Table 2, Fig. 4] Experiment 1 compares a single well-trained and a single overfitting model per architecture/dataset, with no error bars or multiple seeds. Given the fluctuations in anomaly-filter counts shown in Fig. 4(b), the pairwise differences in Table 2 cannot be distinguished from noise. Reporting the mean and standard deviation over at least three independent training runs is necessary to support the claim that anomaly filters "increase in overfitting models."
  4. [Section 4.2, Table 3] The outlier-gradient result is expected from the definition of outliers: samples are labeled as outliers precisely because they form tiny clusters far from the main body in the 2D PCA-GMM embedding (λ=5). Showing that these points have high gradients, while consistent, does not establish that they cause overfitting; indeed, Table 3 shows the same gap in well-trained models. A causal or at least overfitting-specific test (e.g., comparing the gradient gap across training epochs, or ablating outlier samples from the training set) is needed.
  5. [Section 4, first paragraph; Abstract] The paper excludes larger datasets because "insufficient anomaly filters in the model to substantiate our findings." This selection, combined with the claim of universal applicability "without modifications" (Abstract), raises concerns about cherry-picking. The authors should either report results on a larger dataset (e.g., a subset of ImageNet) or characterize the conditions under which anomaly filters exist; otherwise the method's scope is overstated.
  6. [Sections 3.3 and 3.4] The core construction assumes that maximizing the CH Index picks the correct number of clusters K and that small clusters in the 2D PCA embedding correspond to semantically meaningful outlier images. No evidence is given for either claim: there is no comparison to alternative clustering methods or to human-annotated pattern counts, and no validation that the 2D projection preserves the structure relevant to overfitting. Because these assumptions underlie the definition of anomaly filters in every experiment, they must be tested rather than asserted.
minor comments (6)
  1. [Section 3.1] The notation D_l ∈ R^{Batch×C×2} is ambiguous: it is unclear whether PCA is applied separately to each channel's feature maps or to the concatenated matrix; please clarify the exact preprocessing steps.
  2. [Table 1] The hyperparameters λ, α, β, and θ are not defined in the table caption; they are only explained later in Section 4, so the table is hard to interpret in isolation.
  3. [Table 4 caption] The phrase "verification accuracy" should be "validation accuracy" for consistency with the text.
  4. [Section 4.2] The batch size is never reported, although λ=5 is an absolute cluster-size threshold; without knowing the batch size, the threshold has no clear semantic meaning.
  5. [Abstract and Section 1] The sentence "We find that the presence of the anomaly filters suggests potential overfitting" is phrased as a finding, but the paper itself treats it as a hypothesis; the wording should be aligned with the actual experimental evidence.
  6. [Section 4.3] The sentence "If an anomaly occurs on the ReLU or maxpool layer, we simply mask the corresponding upper convolutional filter" is unclear about what "anomaly occurs" means and how the layer type is determined; please specify the rule.

Circularity Check

0 steps flagged · score 2.0 of 10

No by-construction circularity: anomaly-filter counts, gradient gaps, and pruning outcomes are empirical measurements; the only self-citation is non-load-bearing.

full rationale

The paper's claimed derivation chain is empirical rather than deductive: it clusters per-filter feature maps with GMM, defines anomaly filters via unbalanced clusters, high CH index, and activation thresholds, then observes that anomaly-filter counts rise in overfitting models, that outlier samples have larger gradients, and that masking anomaly filters tends to improve validation accuracy. None of these steps is a mathematical consequence of the definitions. The anomaly-filter definition in Section 3.4 does not reference overfitting, so the counts in Table 2 are independent measurements. Experiment 2's gradient comparison is also empirical: samples in small GMM clusters are not by construction high-loss or high-gradient samples, so the observed gradient gap is not entailed by the clustering definition. Experiment 3 is a direct intervention whose outcome is measured on validation accuracy. The only self-citation is [17] in Related Work, which is not load-bearing. Concerns about per-experiment hyperparameters (Table 1: λ=100/5/50, β=1/1.2, θ=0.2/0.5) and the explicit statement in Section 4.3 that 'stricter rules have been applied' are reproducibility and construct-validity issues, not circular reductions: the paper does not define 'anomaly filter' in terms of the outcome it predicts. Similarly, Table 4's LeNet-5 rows (validation accuracy up in 1/3 and 0/1 cases) weaken the generalization claim but do not make it circular. Therefore no step reduces by construction to its inputs; the score of 2 reflects only a minor, non-load-bearing self-citation.

Assumptions & free parameters 5 free parameters · 4 assumptions · 2 invented entities

The central claims rest on several hand-set thresholds and modeling choices (PCA-to-2D, GMM, CH-index maximization for K) that are changed between experiments, plus a conceptual category (anomaly filter) defined by those thresholds and then tested with the same thresholds.

free parameters (5)
  • lambda (small cluster size threshold) = 100 (Exp1), 5 (Exp2), 50 (Exp3)
    Threshold for a cluster to be counted as small/anomaly cluster; varies across experiments and directly controls which filters are called anomaly filters.
  • alpha (unbalanced distribution threshold) = 0.2 (Exp1, Exp3)
    A filter is unbalanced if the number of anomaly clusters exceeds alpha times the total cluster number.
  • beta (CH index multiplier) = 1 (Exp1), 1.2 (Exp3)
    A filter meets the high-CH criterion if its CH index exceeds beta times the layer average; tuned, with a stricter setting in the pruning experiment.
  • theta (activation threshold) = 0.2 (Exp1), 0.5 (Exp3)
    Filters with activation less than theta times the layer average are excluded; value changed per experiment.
  • K (number of GMM clusters) = not specified; selected dynamically by max CH index over an unspecified range
    The paper says 'determine an approximate range' but never gives the range or the exact selection algorithm, so K is effectively a free choice.
assumptions (4)
  • domain assumption Feature maps of each filter, after PCA to 2 dimensions, are generated by a mixture of K Gaussians.
    Section 3.1 assumes the GMM generative model on PCA-reduced feature maps; no validation that 2D PCA preserves the structure needed.
  • ad hoc to paper The K maximizing the CH Index approximates the number of patterns a filter has learned.
    Section 3.3 introduces this without empirical or theoretical support.
  • domain assumption Calinski-Harabasz Index is a suitable unsupervised measure of filter clustering quality for anomaly detection.
    Section 3.2 uses CH index as the sole metric; its behavior with outlier points is well known, but the thresholding is ad hoc.
  • ad hoc to paper Small clusters in the 2D clustering correspond to semantically meaningful outlier images whose gradients drive overfitting.
    Section 4.2 equates cluster outliers with 'outlier samples' without external validation that these are the samples that cause overfitting.
invented entities (2)
  • anomaly filter
    purpose: A filter whose feature-map clustering shows unbalanced class distribution, high CH index, and sufficient activation; proposed as an indicator and cause of overfitting.
    No external falsifiable prediction is provided; the entity is defined by the paper's own thresholds and tested only on the same setup that defines it.
  • outlier samples
    purpose: Images whose feature maps fall in clusters of at most lambda points; claimed to contribute to overfitting via large gradients.
    The label is an artifact of the same clustering pipeline; no independent measure of 'outlierness' is used.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Explaining Model Overfitting in CNNs via GMM Clustering." pith.science (2026). https://pith.science/paper/4T3Q6Q4I

@misc{pith2026241210457,
  author       = {Pith},
  title        = {Pith review of: Explaining Model Overfitting in CNNs via GMM Clustering},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4T3Q6Q4I}},
  note         = {Machine review of arXiv:2412.10457}
}
read the original abstract

Convolutional Neural Networks (CNNs) have demonstrated remarkable prowess in the field of computer vision. However, their opaque decision-making processes pose significant challenges for practical applications. In this study, we provide quantitative metrics for assessing CNN filters by clustering the feature maps corresponding to individual filters in the model via Gaussian Mixture Model (GMM). By analyzing the clustering results, we screen out some anomaly filters associated with outlier samples. We further analyze the relationship between the anomaly filters and model overfitting, proposing three hypotheses. This method is universally applicable across diverse CNN architectures without modifications, as evidenced by its successful application to models like AlexNet and LeNet-5. We present three meticulously designed experiments demonstrating our hypotheses from the perspectives of model behavior, dataset characteristics, and filter impacts. Through this work, we offer a novel perspective for evaluating the CNN performance and gain new insights into the operational behavior of model overfitting.

Figures

Figures reproduced from arXiv: 2412.10457 by the authors.

Figure 1
Figure 1. Given a pre-trained model, we cluster all the feature maps corresponding to the individual filter through the Gaussian Mixture Model. ∙ We provide fine-grained interpretation aligned with the CNN’s nature. We offer the explanations of CNNs at the filter level, in harmony with the inherent char￾acteristics of CNNs. Leveraging unsupervised clus￾tering, our method does not rely on predefined input features or prior hum… view at source ↗
Figure 2
Figure 2. Visualization the clustering results. Each data point corresponds to one feature map generated by one input sample. There are normal cases where data points are evenly distributed and rare cases where outlier points occur. The filter corresponding to the rare type of clustering result is the anomaly filter. results for different components of neural networks in a unified way. Feature Attribution: While saliency maps… view at source ↗
Figure 3
Figure 3. We cluster the feature maps corresponding to the individual filter respectively, i.e., each data point in the clustering results corresponds to one feature map. Different colors represent different filters. Meanwhile, We categorize the clustering results into normal/rare cases. In this work, we mainly focus on the rare case scenario as marked in the red box. 3. Method In this section, we aim to evaluate and interpre… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: The training curve and the number of anomaly filters for a simple CNN in the dataset of CIFAR-10. Obvious overfitting occurs for there is a drop in the accuracy curve and an increase in the loss curve. The number of anomaly filters curve shows a similar trend to the lo…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 7 canonical work pages

  1. [1]

    Rudin, Stop explaining black box machine learning models for highstakesdecisionsanduseinterpretablemodelsinstead, NatMach Intell 1 (2019) 206–215

    C. Rudin, Stop explaining black box machine learning models for highstakesdecisionsanduseinterpretablemodelsinstead, NatMach Intell 1 (2019) 206–215. URL: https://www.nature.com/articles/ s42256-019-0048-x. doi:10.1038/s42256-019-0048-x

    doi:10.1038/s42256-019-0048-x 2019

  2. [2]

    Rudin, C

    C. Rudin, C. Chen, Z. Chen, H. Huang, L. Semenova, C. Zhong, Interpretable Machine Learning: Fundamental Principles and 10 Grand Challenges, 2021. URL: http://arxiv.org/abs/2103.11251, arXiv:2103.11251 [cs, stat]

    arXiv 2021

  3. [3]

    Y. Zhu, J. Ma, C. Yuan, X. Zhu, Interpretable learning based Dy- namic Graph Convolutional Networks for Alzheimer’s Disease anal- ysis, InformationFusion77(2022)53–61.URL: https://linkinghub. elsevier.com/retrieve/pii/S1566253521001548.doi: 10.1016/j.inffus. 2021.07.013

    doi:10.1016/j.inffus 2022

  4. [4]

    J.You,J.Leskovec,K.He,S.Xie,Graphstructureofneuralnetworks, in: H. D. III, A. Singh (Eds.), Proceedings of the 37th International Conference on Machine Learning, volume 119 ofProceedings of Machine Learning Research, PMLR, 2020, pp. 10881–10891. URL: https://proceedings.mlr.press/v119/you20b.html

    work page 2020

  5. [5]

    A Survey on Neural Network Interpretability

    Y. Zhang, P. Tiňo, A. Leonardis, K. Tang, A Survey on Neural Network Interpretability, IEEE Trans. Emerg. Top. Comput. Intell. 5 (2021) 726–742. URL: http://arxiv.org/abs/2012.14261. doi:10. 1109/TETCI.2021.3100641, arXiv:2012.14261 [cs]

    work page Pith review arXiv 2021

  6. [6]

    L. H. Gilpin, D. Bau, B. Z. Yuan, A. Bajwa, M. Specter, L. Ka- gal, Explaining Explanations: An Overview of Interpretability of Machine Learning, 2019. URL: http://arxiv.org/abs/1806.00069, arXiv:1806.00069 [cs, stat]

    arXiv 2019

  7. [7]

    URL: https://linkinghub.elsevier.com/retrieve/pii/ S1051200417302385

    G.Montavon,W.Samek,K.-R.Müller, Methodsforinterpretingand understanding deep neural networks, Digital Signal Processing 73 (2018) 1–15. URL: https://linkinghub.elsevier.com/retrieve/pii/ S1051200417302385. doi:10.1016/j.dsp.2017.10.011

    doi:10.1016/j.dsp.2017.10.011 2018

  8. [8]

    URL:http://arxiv.org/abs/1602

    A.Nguyen,J.Yosinski,J.Clune,MultifacetedFeatureVisualization: UncoveringtheDifferentTypesofFeaturesLearnedByEachNeuron in Deep Neural Networks, 2016. URL:http://arxiv.org/abs/1602. 03616, arXiv:1602.03616 [cs]

    arXiv 2016

Show all 26 references
  1. [9]

    Athiwaratkun, K

    B. Athiwaratkun, K. Kang, Feature Representation in Convolu- tional Neural Networks, 2015. URL: http://arxiv.org/abs/1507. 02313. doi:10.48550/arXiv.1507.02313, arXiv:1507.02313 [cs]

    2015 doi arXiv

  2. [10]

    Dosovitskiy, T

    A. Dosovitskiy, T. Brox, Inverting Visual Representations with Convolutional Networks, in: 2016 IEEE Conference on Computer VisionandPatternRecognition(CVPR),IEEE,LasVegas,NV,USA, 2016, pp. 4829–4837. URL: http://ieeexplore.ieee.org/document/ 7780891/. doi:10.1109/CVPR.2016.522

    2016 doi

  3. [11]

    Zhang, J

    X. Zhang, J. Xu, J. Yang, L. Chen, H. Zhou, X. Liu, H. Li, T. Lin, Y. Ying, Understanding the learning mechanism of convolutional neural networks in spectral analysis, Analytica Chimica Acta 1119 (2020)41–51.URL: https://linkinghub.elsevier.com/retrieve/pii/ S0003267020303767....

    2020 doi

  4. [12]

    Zhang, X

    Q. Zhang, X. Wang, R. Cao, Y. N. Wu, F. Shi, S.-C. Zhu, Extraction of an Explanatory Graph to Interpret a CNN, IEEE Trans. Pattern Anal.Mach.Intell.43(2021)3863–3877.URL: https://ieeexplore. ieee.org/document/9086075/. doi:10.1109/TPAMI.2020.2992207

    2021

  5. [13]

    Girish, V

    D. Girish, V. Singh, A. L. Ralescu, Unsupervised clustering based understanding of cnn., in: CVPR Workshops, 2019, pp. 9–11

    2019

  6. [14]

    B. Zhou, A. Khosla, A. Lapedriza, A. Oliva, A. Torralba, Learning Deep Features for Discriminative Localization, in: 2016 IEEE Con- ferenceonComputerVisionandPatternRecognition(CVPR),IEEE, LasVegas,NV,USA,2016,pp.2921–2929.URL: http://ieeexplore. ieee.org/document/7780688/. do...

    2016

  7. [15]

    R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, D. Batra, Grad-CAM: Visual Explanations from Deep Net- works via Gradient-based Localization, Int J Comput Vis 128 (2020) 336–359. URL: http://arxiv.org/abs/1610.02391. doi: 10. 1007/s11263-019-01228-7, arXiv:1610.0...

    2020 arXiv

  8. [16]

    S. Bach, A. Binder, G. Montavon, F. Klauschen, K.-R. Müller, W. Samek, On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation, PLoS ONE 10 (2015) e0130140. URL:https://dx.plos.org/10.1371/journal.pone. 0130140. doi:10.1371/journal.po...

    2015 doi

  9. [17]

    H. Dou, F. Shen, J. Zhao, X. Mu, Understanding neural net- work through neuron level visualization, Neural Networks 168 (2023) 484–495. URL: https://linkinghub.elsevier.com/retrieve/ pii/S0893608023005269. doi:10.1016/j.neunet.2023.09.030

    2023 doi

  10. [18]

    Why Should I Trust You?

    M. T. Ribeiro, S. Singh, C. Guestrin, "Why Should I Trust You?": Explaining the Predictions of Any Classifier, 2016. URL:http:// arxiv.org/abs/1602.04938, arXiv:1602.04938 [cs, stat]

    2016 arXiv

  11. [19]

    S. M. Lundberg, S.-I. Lee, A unified approach to interpreting model predictions, Advances in neural information processing systems 30 (2017)

    2017

  12. [20]

    B. Kim, M. Wattenberg, J. Gilmer, C. Cai, J. Wexler, F. Viegas, R. Sayres, Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV), 2018. URL:http: //arxiv.org/abs/1711.11279, arXiv:1711.11279 [stat]

    2018 arXiv

  13. [21]

    Zhang, Y

    Q. Zhang, Y. N. Wu, S.-C. Zhu, Interpretable Convolutional Neural Networks, 2018. URL: http://arxiv.org/abs/1710.00935, arXiv:1710.00935 [cs]

    2018 arXiv

  14. [22]

    Calinski, J

    T. Calinski, J. Harabasz, A dendrite method for cluster analysis, Comm. in Stats. - Theory & Methods 3 (1974) 1–27. URL: http://www.tandfonline.com/doi/abs/10.1080/03610927408827101. doi:10.1080/03610927408827101

    1974 doi

  15. [23]

    A.P.Dempster,N.M.Laird,D.B.Rubin, Maximumlikelihoodfrom incomplete data via the em algorithm, Journal of the royal statistical society: series B (methodological) 39 (1977) 1–22

    1977

  16. [24]

    Krizhevsky, I

    A. Krizhevsky, I. Sutskever, G. E. Hinton, ImageNet classification with deep convolutional neural networks, Commun. ACM 60 (2017) 84–90. URL: https://dl.acm.org/doi/10.1145/3065386. doi:10.1145/ 3065386

    2017 doi

  17. [25]

    Lecun, L

    Y. Lecun, L. Bottou, Y. Bengio, P. Haffner, Gradient-based learning applied to document recognition, Proc. IEEE 86 (1998) 2278–2324. URL: http://ieeexplore.ieee.org/document/726791/. doi:10.1109/5. 726791

    1998 doi

  18. [26]

    First Author et al.:Preprint submitted to Elsevier Page 9 of 9

    N.S.Keskar,D.Mudigere,J.Nocedal,M.Smelyanskiy,P.T.P.Tang, On large-batch training for deep learning: Generalization gap and sharp minima, arXiv preprint arXiv:1609.04836 (2016). First Author et al.:Preprint submitted to Elsevier Page 9 of 9

    2016 arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.